Data processing method and device, and vehicle

Abstract

The application provides a data processing method and device and a vehicle, and belongs to the technical field of computers. The method comprises the following steps: acquiring a plurality of image data, and performing entity identification based on the plurality of image data to obtain a plurality of entities and an initial confidence corresponding to each entity; for each entity, in response to the initial confidence corresponding to the entity being less than a first threshold, performing entity identification again after performing enhancement processing on the image data corresponding to the entity in the plurality of image data, to obtain an enhanced confidence corresponding to the entity; determining a comprehensive confidence corresponding to the entity based on the initial confidence and the enhanced confidence corresponding to the entity; and performing data desensitization on the entity based on the comprehensive confidence corresponding to the entity, the first threshold and a preset parameter, to generate desensitized data. In this way, the accuracy of entity identification can be improved, and the generated desensitized data has high usability.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a data processing method, apparatus, and vehicle. Background Technology

[0002] With the rapid development of intelligent connected vehicle technology, image data (such as images, videos, and point clouds) generated by vehicles has become a core resource for the iteration of autonomous driving models and the updating of high-precision maps. However, image data often contains sensitive personal information such as faces, license plates, and house numbers. In order to comply with the regulations on the protection of sensitive personal information, it is necessary to de-identify the sensitive personal information in the image data to generate de-identified data.

[0003] In related technologies, existing data processing methods mainly rely on target detection models to identify multiple image data, obtain entities and their corresponding confidence scores, and perform desensitization processing based on the confidence scores of the entities to generate desensitized data. However, in complex scenarios such as insufficient lighting, bad weather, target occlusion, or excessive distance, the confidence scores of some image data identified by this method are generally low. This can easily lead to the misclassification of a large number of low-confidence non-sensitive entities (such as tree shadows and building textures) as sensitive targets, thus applying the desensitization strategy corresponding to sensitive data to non-sensitive entities. This mis-desensitization phenomenon can seriously damage the integrity and usability of image data. Summary of the Invention

[0004] The purpose of this application is to provide a data processing method, apparatus, and vehicle to avoid erroneous data anonymization and improve the accuracy of data anonymization. The specific technical solution is as follows: In a first aspect of this application, a data processing method is provided, the method comprising: Multiple image data are acquired, and entity recognition is performed based on the multiple image data to obtain multiple entities and the initial confidence level corresponding to each entity; For each entity, in response to the initial confidence level corresponding to the entity being less than a first threshold, the image data corresponding to the entity in the multiple image data is enhanced and then entity recognition is performed again to obtain the enhanced confidence level corresponding to the entity. Based on the initial confidence level and the enhanced confidence level corresponding to the entity, the overall confidence level corresponding to the entity is determined; Based on the comprehensive confidence level of the entity, the first threshold, and preset parameters, the entity is de-identified to generate de-identified data.

[0005] In a second aspect of this application, a data processing apparatus is also provided, the apparatus comprising: An entity recognition module is used to acquire multiple image data and perform entity recognition based on the multiple image data to obtain multiple entities and the initial confidence level corresponding to each entity; An enhanced confidence determination module is used to, for each entity, in response to the initial confidence corresponding to the entity being less than a first threshold, perform enhanced processing on the image data corresponding to the entity in the multiple image data and then re-perform entity recognition to obtain the enhanced confidence corresponding to the entity; The comprehensive confidence level determination module is used to determine the comprehensive confidence level of the entity based on the initial confidence level and the enhanced confidence level corresponding to the entity. The de-identified data generation module is used to de-identify the entity based on the comprehensive confidence level corresponding to the entity, the first threshold, and preset parameters, and generate de-identified data.

[0006] In a third aspect of the embodiments of this application, a vehicle is also provided, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the data processing method described in any one of the first aspects above.

[0007] In a fourth aspect of the embodiments of this application, a storage medium is also provided, wherein the storage medium stores instructions that, when run on a computer, cause the computer to perform any of the data processing methods described in the first aspect above.

[0008] In a fifth aspect of the embodiments of this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the data processing methods described in the first aspect above.

[0009] The technical solution provided in this application involves acquiring multiple image data sets and performing entity recognition based on these image data sets to obtain multiple entities and an initial confidence level for each entity. For each entity, in response to an initial confidence level less than a first threshold, the image data corresponding to the entity in the multiple image data sets is enhanced and entity recognition is performed again to obtain an enhanced confidence level for the entity. Based on the initial and enhanced confidence levels of the entity, a comprehensive confidence level is determined. Based on the comprehensive confidence level of the entity, the first threshold, and preset parameters, data anonymization is performed on the entity to generate anonymized data. By enhancing the image data corresponding to entities with an initial confidence level below the first threshold and then re-identifying the entities to obtain a comprehensive confidence level, the entity data can be desensitized to generate desensitized data. This improves the accuracy of entity recognition and makes the generated desensitized data more usable. It solves the technical problem that in complex scenarios such as insufficient lighting, bad weather, target occlusion, or excessive distance, the confidence scores of multiple image data are generally low. This makes it easy to misidentify a large number of low-confidence non-sensitive entities (such as tree shadows and building textures) as sensitive targets, and thus apply the desensitization strategy corresponding to sensitive data to non-sensitive entities. This false desensitization phenomenon will seriously damage the integrity and usability of multiple image data. Attached Figure Description

[0010] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] One or more embodiments are illustrated by way of example with reference numerals in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0013] Figure 1 A schematic diagram illustrating the implementation flow of a data processing method provided in this application embodiment; Figure 2 A schematic diagram illustrating the implementation flow of another data processing method provided in this application embodiment; Figure 3 A schematic diagram illustrating the implementation flow of another data processing method provided in this application embodiment; Figure 4A schematic diagram illustrating the implementation process of a method for determining a desensitization strategy provided in an embodiment of this application; Figure 5 A schematic diagram illustrating the implementation process of another method for determining a desensitization strategy provided in this application embodiment; Figure 6 This is a schematic diagram of the structure of a data processing device provided in an embodiment of this application; Figure 7 This is a structural schematic diagram of a vehicle provided in an embodiment of this application. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] The following disclosure provides numerous different embodiments or examples for implementing various structures of this application. To simplify the disclosure, specific examples of components and arrangements are described below. These are merely examples and are not intended to limit the scope of this application. Furthermore, reference numerals and / or letters may be repeated in different examples. Such repetition is for simplification and clarity and does not in itself indicate a relationship between the various embodiments and / or arrangements discussed.

[0016] To address the problem that existing technologies often produce low confidence scores when identifying multiple image data sets in complex scenarios such as insufficient lighting, inclement weather, target occlusion, or excessive distance, easily misclassifying numerous low-confidence non-sensitive entities (such as tree shadows and building textures) as sensitive targets, thus applying desensitization strategies corresponding to sensitive data to non-sensitive entities, and this mis-desensitization phenomenon severely compromises the integrity and usability of multiple image data sets, this application provides a data processing method, apparatus, and vehicle. The method involves acquiring multiple image data sets and performing entity recognition based on these sets to obtain multiple entities and their initial confidence scores. For each entity, in response to an initial confidence score less than a first threshold, the image data corresponding to the entity in the multiple image data sets is enhanced, and entity recognition is performed again to obtain an enhanced confidence score. Based on the initial and enhanced confidence scores, a comprehensive confidence score is determined. Finally, based on the comprehensive confidence score, the first threshold, and preset parameters, the entity is desensitized to generate desensitized data. By enhancing the image data corresponding to entities with an initial confidence level below the first threshold and then re-identifying the entities to obtain a comprehensive confidence level, the entity data can be desensitized to generate desensitized data. This can improve the accuracy of entity recognition and make the generated desensitized data more usable.

[0017] like Figure 1 The diagram shown is a schematic representation of an implementation flow of a data processing method provided in this application, which may specifically include the following steps: S101, acquire multiple image data, and perform entity recognition based on the multiple image data to obtain multiple entities and the initial confidence level corresponding to each entity.

[0018] The aforementioned image data refers to the raw data collected in real time by vehicle-mounted sensors (such as forward / surround view cameras, lidar, millimeter-wave radar, cabin cameras, etc.) during vehicle operation. The data may include image frames, video streams, 3D point clouds, etc., and this application embodiment does not limit the data in this way.

[0019] The aforementioned entities refer to detectable targets (i.e. entities) in multiple image data, which may include entities corresponding to categories such as pedestrians, vehicles, faces, license plates, and traffic signs.

[0020] The initial confidence level mentioned above is used to characterize the probability of the category corresponding to the entity and its location accuracy, in order to measure the accuracy of the current entity recognition result, such as 0.5 or 0.6.

[0021] In this embodiment of the application, multiple image data are acquired, and entity recognition is performed based on the multiple image data to obtain multiple entities and the initial confidence level corresponding to each entity.

[0022] Specifically, an in-vehicle AI model (such as the YOLOv5s multi-task detection model) can be used to perform multi-task detection (such as entity recognition and entity annotation) on multiple image data, thereby obtaining the category, initial confidence level and bounding box information of the region corresponding to each entity. This application embodiment does not limit this.

[0023] S102, for each entity, in response to the initial confidence level corresponding to the entity being less than the first threshold, the image data corresponding to the entity in multiple image data is enhanced and then entity recognition is performed again to obtain the enhanced confidence level corresponding to the entity.

[0024] In this embodiment, for each entity, in response to the initial confidence level corresponding to the entity being less than a first threshold, the image data corresponding to the entity in multiple image data sets is enhanced and then entity recognition is performed again to obtain the enhanced confidence level corresponding to the entity. The first threshold can be understood as a low-confidence threshold, used to determine the accuracy of the recognition result corresponding to the entity, such as 0.3. The image data corresponding to the entity refers to the image data containing the entity within the image data containing the entity.

[0025] S103, Based on the initial confidence level and enhanced confidence level corresponding to the entity, determine the comprehensive confidence level corresponding to the entity.

[0026] In this embodiment, the overall confidence level of an entity is determined based on the initial confidence level and the enhanced confidence level corresponding to the entity. The overall confidence level refers to the evaluation score of the final reliability of entity recognition, such as 0.5.

[0027] S104. Based on the comprehensive confidence level of the entity, the first threshold, and preset parameters, perform data anonymization on the entity and generate anonymized data.

[0028] In this embodiment, based on the entity's comprehensive confidence level, a first threshold, and preset parameters, data anonymization is performed on the entity to generate anonymized data. Data anonymization refers to applying irreversible transformation processing to identified sensitive entities or their corresponding data to remove or mask their identity information while preserving their geometric, structural, and semantic information as much as possible. The anonymized data is used to train the autonomous driving model for vehicles, while meeting privacy compliance requirements.

[0029] It should be noted that the output format of the de-identified data can be a "controlled data package," which can contain three core types of content: First, de-identified multimodal data, i.e., image, point cloud, and other data processed by the above strategies, where the original sensitive information has been irreversibly masked; second, annotation files, which can be in XML / JSON format, used to record the identifier, category, bounding box coordinates, overall confidence level, and de-identification strategy type of valid entities; and third, enhancement operation logs, used to record the enhancement operator type, enhancement region (i.e., the region corresponding to the entity that needs enhancement processing), and initial / enhanced / overall confidence level of low-confidence entities, providing data support for cloud auditing and model feedback optimization.

[0030] Based on the above description of the technical solution provided in the embodiments of this application, multiple image data are acquired, and entity recognition is performed based on the multiple image data to obtain multiple entities and an initial confidence level corresponding to each entity; for each entity, in response to the initial confidence level corresponding to the entity being less than a first threshold, the image data corresponding to the entity in the multiple image data is enhanced and entity recognition is performed again to obtain the enhanced confidence level corresponding to the entity; based on the initial confidence level and enhanced confidence level corresponding to the entity, the comprehensive confidence level corresponding to the entity is determined; based on the comprehensive confidence level corresponding to the entity, the first threshold, and preset parameters, data anonymization is performed on the entity to generate anonymized data. By enhancing the image data corresponding to entities with an initial confidence level below the first threshold and then re-identifying the entities to obtain a comprehensive confidence level, the entity data can be desensitized to generate desensitized data. This improves the accuracy of entity recognition and makes the generated desensitized data more usable. It solves the technical problem that in complex scenarios such as insufficient lighting, bad weather, target occlusion, or excessive distance, the confidence scores of multiple image data are generally low. This makes it easy to misidentify a large number of low-confidence non-sensitive entities (such as tree shadows and building textures) as sensitive targets, thus applying the desensitization strategy corresponding to sensitive data to non-sensitive entities. This false desensitization phenomenon will seriously damage the integrity and usability of multiple image data.

[0031] like Figure 2 The diagram shown illustrates the implementation flow of another data processing method provided in this application, which may specifically include the following: S201, acquire multiple image data, and perform entity recognition based on the multiple image data to obtain multiple entities and the initial confidence level corresponding to each entity.

[0032] In this embodiment of the application, this step is similar to step S101 above, and will not be described in detail here.

[0033] S202, for each entity, in response to the initial confidence level corresponding to the entity being less than the first threshold, the image data corresponding to the entity in the multiple image data is enhanced and then entity recognition is performed again to obtain the enhanced confidence level corresponding to the entity.

[0034] The aforementioned enhancement processing refers to a lightweight real-time image and data processing operation performed on entities with an initial confidence level less than a first threshold in the region of the original image data corresponding to the entity (such as a cropped image patch) across multiple image datasets.

[0035] In this embodiment of the application, for each entity, in response to the initial confidence level corresponding to the entity being less than a first threshold, the image data corresponding to the entity in multiple image data is enhanced and then entity recognition is performed again to obtain the enhanced confidence level corresponding to the entity.

[0036] The first threshold can be obtained from a preset first threshold mapping table based on the category corresponding to the entity. The category corresponding to the entity can include one of the following: face, license plate, vehicle, pedestrian, traffic sign, etc. The preset first threshold mapping table is a pre-defined table that stores low-confidence thresholds corresponding to different categories.

[0037] For example, the first threshold is set as follows: for the category of human face, the corresponding low confidence threshold is 0.5; for the category of license plate, the corresponding low confidence threshold is 0.3; and for the category of vehicle, the corresponding low confidence threshold is 0.1.

[0038] The process of enhancing the image data corresponding to entities in multiple image datasets and then re-performing entity recognition to obtain the enhanced confidence score for each entity can specifically include the following steps: Step 1: Obtain at least one of the environmental information and attributes corresponding to the entity.

[0039] In this embodiment, at least one of the environmental information and attributes corresponding to the entity is obtained. Specifically, at least one of the environmental information and attributes corresponding to the entity can be obtained from multiple image data. The environmental information corresponding to the entity refers to the macroscopic scene state of the vehicle when multiple image data are collected. This information can be obtained through an onboard communication and perception module (such as a CAN bus or sensor data fusion system). For example, the environment in the environmental information can be determined to be nighttime using timestamps and light intensity sensors; the environment in the environmental information can be determined to be rainy or foggy using rain sensors and image feature analysis; and the location of the environmental information can be determined to be urban roads or highways using GPS and high-precision map data. The attributes corresponding to the entity refer to the entity's own characteristics, such as the entity's size and motion state. For example, for image-based perception data, the entity's attribute can be determined as "small target" (easily blurred) or "large target" based on the ratio of the pixel area of ​​the entity's bounding box to the total image area; the existence of "dynamic blur" can be determined based on the degree of motion blur of the image block. This embodiment does not limit this aspect.

[0040] Step 2: Determine the target enhancement strategy from the preset strategy table based on at least one of the environmental information and attributes.

[0041] In this embodiment, a target enhancement strategy is determined from a preset strategy table based on at least one of environmental information and attributes. The preset strategy table contains the mapping relationship between environmental information and / or attributes and enhancement strategies. The enhancement strategy is used to enhance the data corresponding to the entity and can be a CLAHE operator, a Retinex dehazing operator, or a lightweight super-resolution reconstruction operator; this embodiment does not limit the specific type of enhancement strategy.

[0042] For example, if the environment in which the vehicle is located is at night and the attribute is small target, the target enhancement strategy includes the CLAHE operator; if the environment information is rainy or foggy weather, the target enhancement strategy includes the Retinex defogging operator.

[0043] Step 3: Based on the target enhancement strategy, enhance the image data corresponding to entities in multiple image data to obtain enhanced image data.

[0044] In this embodiment, image data corresponding to entities in multiple image datasets are enhanced using a target enhancement strategy to obtain enhanced image data. Enhanced image data refers to local data blocks obtained after enhancing the image data corresponding to entities. Depending on the shape of the multiple image datasets, this can correspond to image sub-blocks, point cloud blocks, etc. The image data corresponding to entities can be local data blocks cropped from the original multiple image datasets after appropriately expanding the bounding box corresponding to the entity (e.g., expanding the bounding box area by 10%-30% in all directions). This embodiment does not limit this approach.

[0045] Step 4: Perform entity recognition on the enhanced image data to obtain the enhanced confidence scores of the entities.

[0046] In this embodiment, entity recognition is performed on the enhanced image data to obtain the enhancement confidence score corresponding to each entity. Specifically, an in-vehicle AI model (such as the YOLOv5s multi-task detection model) can be used to recognize the enhanced image data and obtain the enhancement confidence score corresponding to each entity.

[0047] S203, obtain the preset attenuation factor.

[0048] In this embodiment, a preset attenuation factor is obtained. The preset attenuation factor is a pre-set weighting parameter used to adjust the contribution ratio of the initial confidence level and the enhanced confidence level of the entity to the overall confidence level.

[0049] S204, subtract the preset first value from the preset attenuation factor to obtain the first calculation result.

[0050] In this embodiment of the application, a first calculation result is obtained by subtracting a preset first value from a preset attenuation factor.

[0051] For example, if the preset first value is 1 and the attenuation factor is 0.5, then subtracting the preset first value from the preset attenuation factor will result in the first operation result of 1-0.5=0.5.

[0052] S205, multiply the preset attenuation factor by the initial confidence level to obtain the second calculation result, and multiply the first calculation result by the enhanced confidence level to obtain the third calculation result.

[0053] In this embodiment of the application, a preset attenuation factor is multiplied by an initial confidence level to obtain a second calculation result, and a first calculation result is multiplied by an enhanced confidence level to obtain a third calculation result.

[0054] For example, if the preset attenuation factor is 0.5 and the initial confidence level is 0.5, multiplying the preset attenuation factor by the initial confidence level yields a second result of 0.25, the first result is 0.5, and the enhanced confidence level is 0.6, multiplying the first result by the enhanced confidence level yields a third result of 0.3.

[0055] S206, add the first operation result and the third operation result to obtain the overall confidence level corresponding to the entity.

[0056] In this embodiment of the application, the first calculation result and the third calculation result are added together to obtain the overall confidence level corresponding to the entity.

[0057] For example, in the example above, the result of the second operation is known to be 0.25 and the result of the third operation is 0.3. Adding the result of the first operation to the result of the third operation yields a comprehensive confidence level of 0.55 for the entity.

[0058] Specifically, when the preset first value is 1, the preset attenuation factor, the initial confidence level corresponding to the entity, and the enhanced confidence level can be input into the comprehensive confidence level formula to calculate the comprehensive confidence level corresponding to the entity. The comprehensive confidence level formula is as follows: ; in, The overall confidence level corresponding to the entity. As a preset attenuation factor, The initial confidence level for the entity. Enhance the confidence level corresponding to the entity.

[0059] For example, if the preset attenuation factor is 0.3, the initial confidence level of the entity is 0.25, the enhanced confidence level of the entity is 0.65, then the overall confidence level of the entity is 0.53.

[0060] S207, based on the comprehensive confidence level of the entity, the first threshold and preset parameters, perform data anonymization on the entity and generate anonymized data.

[0061] In this embodiment of the application, data anonymization of an entity can be performed based on the entity's comprehensive confidence level, a first threshold, and preset parameters to generate anonymized data.

[0062] For preset parameters, the balance factor corresponding to the category can be obtained from the preset factor mapping table based on the category of the entity, and used as the preset parameter.

[0063] The preset factor mapping table is a pre-defined table that stores balance factors corresponding to different categories. Preset parameters are used to quantify the trade-off between privacy protection strength and data availability; the closer the preset parameter is to 1, the more strictly the de-identification strategy for the entity tends to protect privacy. This application does not limit this aspect.

[0064] For example, the first preset threshold is set to the following categories: face, with a preset parameter of 0.9; license plate, with a preset parameter of 0.8; and vehicle, with a preset parameter of 0.5.

[0065] like Figure 3 The diagram shown illustrates the implementation flow of another data processing method provided in this application, which may specifically include the following: S301, acquire multiple image data, and perform entity recognition based on the multiple image data to obtain multiple entities and the initial confidence level corresponding to each entity.

[0066] In this embodiment of the application, this step is similar to step S101 above, and will not be described in detail here.

[0067] S302, for each entity, in response to the initial confidence level corresponding to the entity being less than the first threshold, the image data corresponding to the entity in the multiple image data is enhanced and then entity recognition is performed again to obtain the enhanced confidence level corresponding to the entity.

[0068] In this embodiment of the application, this step is similar to step S102 above, and will not be described in detail here.

[0069] S303, Based on the initial confidence level and enhanced confidence level corresponding to the entity, determine the comprehensive confidence level corresponding to the entity.

[0070] In this embodiment of the application, this step is similar to step S103 above, and will not be described in detail here.

[0071] S304, obtain the second threshold corresponding to the category of the entity.

[0072] In this embodiment, a second threshold corresponding to the category of the entity is obtained. Specifically, the threshold corresponding to the category can be obtained from a preset second threshold mapping table based on the category of the entity, and used as the second threshold. The preset second threshold mapping table is a pre-defined table containing the correspondence between categories and thresholds. The second threshold can be understood as a high-confidence threshold, used to determine whether the entity recognition result is highly accurate. Different second thresholds can be set for different categories; this embodiment does not limit this.

[0073] S305. Based on the comprehensive confidence level, preset parameters, first threshold and second threshold corresponding to the entity, determine the de-identification strategy corresponding to the entity.

[0074] In this embodiment, the de-identification strategy for an entity can be determined based on the entity's comprehensive confidence level, preset parameters, a first threshold, and a second threshold. The de-identification strategy refers to a specific privacy processing and data availability preservation scheme determined for the entity, which may include whether to de-identify, what de-identification technology to use, the strength of the de-identification, and whether to retain the entity's annotation information. It can be divided into strong de-identification strategies and weak de-identification strategies, but this embodiment does not limit this.

[0075] For details on how to determine the de-identification strategy for an entity based on its comprehensive confidence level, preset parameters, first threshold, and second threshold, please refer to [reference needed]. Figure 4 The method shown. (As illustrated) Figure 4The diagram shown illustrates the implementation flow of a method for determining a desensitization strategy according to an embodiment of this application, which may specifically include the following steps: S401, in response to the fact that the category corresponding to the entity belongs to the preset sensitive class, the product of the preset parameter and the second threshold is used as the first desensitization threshold.

[0076] In this embodiment, in response to the entity's corresponding category belonging to a preset sensitive category, the product of a preset parameter and a second threshold is used as the first desensitization threshold. Specifically, it can be determined whether the entity's corresponding category belongs to a preset sensitive category. If the entity's corresponding category belongs to a preset sensitive category, the product of the preset parameter and the second threshold is used as the first desensitization threshold.

[0077] Among them, the preset sensitive categories refer to a set of categories that are pre-defined and related to personal information or privacy, such as faces, license plate numbers, ID cards, and residential house numbers. This application embodiment does not limit this.

[0078] For example, if the preset sensitive categories include faces, license plate numbers, ID cards, and house numbers, and entity 1 corresponds to the category of trees, while entity 2 corresponds to the category of faces, then the category corresponding to entity 1 does not belong to the preset sensitive categories, while the category corresponding to entity 2 does. Given that the category corresponding to entity 2 belongs to the preset sensitive categories, the preset parameter for entity 2 is 0.9, and the second threshold is 0.7, then the first desensitization threshold is 0.9 × 0.7 = 0.63.

[0079] S402, determine the noise threshold based on preset parameters and the first threshold.

[0080] In this embodiment of the application, the noise threshold is determined based on preset parameters and a first threshold.

[0081] Specifically, the fourth calculation result can be obtained by subtracting the preset second value from the preset parameter, and the fourth calculation result can be multiplied by the first threshold to obtain the noise threshold.

[0082] When the preset second value is 1, the noise threshold is determined based on the preset second value and the first threshold. The noise threshold can be calculated by inputting the preset parameter and the first threshold into the noise threshold formula, which is: ; in, Noise threshold These are preset parameters. This is the first threshold.

[0083] For example, in the example above, the category corresponding to entity 2 belongs to the preset sensitive category, the preset parameter corresponding to entity 2 is 0.9, the first threshold is 0.3, then the noise threshold is (1-0.9)×0.3=0.03.

[0084] S403, based on the comprehensive confidence level of the entity, the first desensitization threshold, and the noise threshold, determine the desensitization strategy for the entity.

[0085] In the embodiments of this application, the desensitization strategy corresponding to an entity can be determined based on the comprehensive confidence level, the first desensitization threshold, and the noise threshold corresponding to the entity.

[0086] For details on how to determine the de-identification strategy for an entity based on its comprehensive confidence level, the first de-identification threshold, and the noise threshold, please refer to [reference needed]. Figure 5 The method shown. (As illustrated) Figure 5 The diagram shown illustrates the implementation flow of another method for determining a desensitization strategy provided in this application, which may specifically include the following steps: S501, in response to the overall confidence level of the entity being greater than the first desensitization threshold, the preset first desensitization strategy is used as the desensitization strategy for the entity.

[0087] In this embodiment, in response to the overall confidence level corresponding to an entity being greater than a first de-identification threshold, a preset first de-identification strategy is used as the de-identification strategy for the entity. Specifically, it can be determined whether the overall confidence level corresponding to the entity is greater than the first de-identification threshold. If the overall confidence level corresponding to the entity is greater than the first de-identification threshold, then the preset first de-identification strategy is used as the de-identification strategy for the entity.

[0088] It should be noted that the preset first desensitization strategy can be understood as a strong desensitization strategy, referring to data transformation operations that completely and irreversibly eliminate the identity identifiability of the entity. This can include: complete masking: completely covering the entity's bounding box area with a solid color (such as black); high-intensity pixelation / mosaic: coarsely processing the pixels within the area, causing complete loss of detail; and deterministic deletion and filling: replacing the pixel values ​​of the area corresponding to the entity with statistical values ​​(such as the mean) of the surrounding background, achieving a visual "erasure".

[0089] For example, if the overall confidence level of an entity is 0.35 and the first de-identification threshold is 0.63, and the overall confidence level of the entity (0.35) is less than the first de-identification threshold (0.63), then the overall confidence level of the entity is not greater than the first de-identification threshold. If the overall confidence level of an entity is 0.8 and the first de-identification threshold is 0.63, and the overall confidence level of the entity (0.8) is greater than the first de-identification threshold (0.63), then the overall confidence level of the entity is greater than the first de-identification threshold.

[0090] S502, in response to the fact that the overall confidence level corresponding to the entity is less than or equal to the first desensitization threshold, the desensitization strategy corresponding to the entity is determined based on the overall confidence level and the noise threshold corresponding to the entity.

[0091] In this embodiment, in response to the overall confidence level corresponding to an entity being less than or equal to a first de-identification threshold, a de-identification strategy for the entity is determined based on the overall confidence level and a noise threshold. Specifically, if the overall confidence level corresponding to an entity is less than or equal to the first de-identification threshold, then a de-identification strategy for the entity is determined based on the overall confidence level and a noise threshold.

[0092] Specifically, determining the de-identification strategy for an entity based on its overall confidence level and noise threshold can be achieved by using a pre-defined second de-identification strategy if the overall confidence level of the entity is greater than the noise threshold. This can include the following steps: Step 1: Determine whether the overall confidence level corresponding to the entity is greater than the noise threshold.

[0093] In this embodiment of the application, it is determined whether the overall confidence level corresponding to the entity is greater than the noise threshold.

[0094] For example, if the overall confidence level of entity 1 is 0.5 and the noise threshold is 0.1, then the overall confidence level of entity 1 is greater than the noise threshold; if the overall confidence level of entity 2 is 0.05 and the noise threshold is 0.1, then the overall confidence level of entity 2 is not greater than the noise threshold.

[0095] Step 2: If the overall confidence level of the entity is greater than the noise threshold, then the preset second desensitization strategy is used as the desensitization strategy for the entity.

[0096] In this embodiment, if the overall confidence level corresponding to an entity is greater than the noise threshold, then a weak desensitization strategy is used as the desensitization strategy for that entity. That is, when the overall confidence level corresponding to an entity is greater than the noise threshold, a preset second desensitization strategy is used as the desensitization strategy for that entity.

[0097] It should be noted that the preset second desensitization strategy can be understood as a weak desensitization strategy. It refers to a data transformation operation that deliberately retains non-sensitive semantic information such as the geometric shape, outline, and texture structure of the entity while removing or obfuscating the identity-identifying features corresponding to the entity. This is used to maximize the usability of multiple image data for subsequent model training while meeting the basic requirements of privacy desensitization.

[0098] The second pre-defined desensitization strategy may include differential privacy noise addition: adding random noise (such as Laplacian noise) that conforms to the definition of differential privacy to the pixel values ​​of the region corresponding to the entity, ensuring that the risk of privacy leakage is bounded, while the overall structure of the image and the entity outline are maintained, and the bounding box and other annotation information can be completely preserved.

[0099] The second pre-defined desensitization strategy may include generative replacement: using a lightweight generative adversarial network to replace entities (such as real faces / license plates) with virtual faces / license plates generated by an intelligent agent that do not correspond to real identities, while maintaining the original pose, lighting and position.

[0100] The second pre-defined desensitization strategy may also include edge-preserving blurring: using edge detection technology, only the interior of the entity is blurred, while its outer contour is strictly preserved.

[0101] In another embodiment of this application, if the overall confidence level of an entity is not greater than the noise threshold, the entity is regarded as noise and there is no need to execute the corresponding desensitization strategy. At the same time, the bounding box of the entity does not need to be labeled.

[0102] The above steps S401~S403 describe a method for determining the desensitization strategy for an entity when the category corresponding to the entity belongs to a preset sensitive category. However, in another embodiment of this application, if the category corresponding to the entity does not belong to a preset sensitive category, then the corresponding desensitization strategy does not need to be executed.

[0103] S306, Perform data desensitization on the entity according to the desensitization strategy corresponding to the entity, and generate desensitized data.

[0104] In this embodiment of the application, the entity is de-identified according to the de-identification strategy corresponding to the entity, and de-identified data is generated.

[0105] It should be noted that after desensitizing the entity according to the corresponding desensitization strategy, the entity's annotation information (such as the category "face" and bounding box) can be retained in the output annotation file, and a "strongly desensitized" mark will be added to ensure that the data is still usable for certain macro-analysis tasks (such as people flow statistics).

[0106] Corresponding to the above method embodiments, this application also provides a data processing apparatus, such as... Figure 6 As shown, the device may include an entity recognition module 601, an enhanced confidence determination module 602, a comprehensive confidence determination module 603, and a de-identified data generation module 604.

[0107] The entity recognition module 601 is used to acquire multiple image data and perform entity recognition based on the multiple image data to obtain multiple entities and the initial confidence level corresponding to each entity; The enhanced confidence determination module 602 is used to perform enhanced processing on the image data corresponding to the entity in multiple image data and then re-identify the entity to obtain the enhanced confidence of the entity for each entity in response to the initial confidence of the entity being less than a first threshold. The comprehensive confidence level determination module 603 is used to determine the comprehensive confidence level of an entity based on the initial confidence level and the enhanced confidence level of the entity. The de-identified data generation module 604 is used to de-identify the entity based on the entity's comprehensive confidence level, first threshold, and preset parameters, and generate de-identified data.

[0108] This application also provides a vehicle, such as... Figure 7 As shown, it includes a processor 701, a communication interface 702, a memory 703, and a communication bus 704, wherein the processor 701, the communication interface 702, and the memory 703 communicate with each other through the communication bus 704. Memory 703 is used to store computer programs; In one embodiment of this application, when the processor 701 executes a program stored in the memory 703, it performs the following steps: Multiple image data sets are acquired, and entity recognition is performed based on these image data sets to obtain multiple entities and their corresponding initial confidence scores. For each entity, in response to the initial confidence score being less than a first threshold, the image data corresponding to the entity in the multiple image data sets is enhanced, and entity recognition is performed again to obtain the enhanced confidence score. Based on the initial and enhanced confidence scores of the entity, the comprehensive confidence score of the entity is determined. Based on the comprehensive confidence score, the first threshold, and preset parameters, the entity data is de-identified to generate de-identified data.

[0109] The communication bus mentioned in the above vehicles can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not indicate that there is only one bus or one type of bus.

[0110] The communication interface is used for communication between the aforementioned vehicle and other devices.

[0111] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0112] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0113] In another embodiment provided in this application, a storage medium is also provided, which stores instructions that, when run on a computer, cause the computer to execute any of the data processing methods described in the above embodiments.

[0114] In another embodiment provided in this application, a computer program product containing instructions is also provided, which, when run on a computer, causes the computer to perform any of the data processing methods described in the above embodiments.

[0115] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a storage medium or transmitted from one storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0116] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0117] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0118] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application are included within the protection scope of this application.

Claims

1. A data processing method, characterized in that, The method includes: Multiple image data are acquired, and entity recognition is performed based on the multiple image data to obtain multiple entities and the initial confidence level corresponding to each entity; For each entity, in response to the initial confidence level corresponding to the entity being less than a first threshold, the image data corresponding to the entity in the multiple image data is enhanced and then entity recognition is performed again to obtain the enhanced confidence level corresponding to the entity. Based on the initial confidence level and the enhanced confidence level corresponding to the entity, the overall confidence level corresponding to the entity is determined; Based on the comprehensive confidence level of the entity, the first threshold, and preset parameters, the entity is de-identified to generate de-identified data.

2. The method according to claim 1, characterized in that, The step of enhancing the image data corresponding to the entity in the plurality of image data and then re-performing entity recognition to obtain the enhanced confidence score corresponding to the entity includes: Obtain at least one of the environmental information and attributes corresponding to the entity; Based on at least one of the environmental information and attributes, a target enhancement strategy is determined from a preset strategy table; Based on the target enhancement strategy, the image data corresponding to the entity in the multiple image data are enhanced to obtain the enhanced image data; Entity recognition is performed on the enhanced image data to obtain the enhanced confidence level corresponding to the entity.

3. The method according to claim 1, characterized in that, The step of determining the overall confidence level of the entity based on the initial confidence level and the enhanced confidence level includes: Obtain the preset attenuation factor; The first calculation result is obtained by subtracting the preset first value from the preset attenuation factor. The preset attenuation factor is multiplied by the initial confidence level to obtain a second calculation result, and the first calculation result is multiplied by the enhanced confidence level to obtain a third calculation result; The first calculation result is added to the third calculation result to obtain the overall confidence level corresponding to the entity.

4. The method according to claim 1, characterized in that, The step of de-identifying the entity based on its comprehensive confidence level, the first threshold, and preset parameters to generate de-identified data includes: Obtain the second threshold corresponding to the category of the entity; Based on the comprehensive confidence level corresponding to the entity, the preset parameters, the first threshold and the second threshold, the de-identification strategy corresponding to the entity is determined; The entity is de-identified according to the de-identification strategy corresponding to the entity, and the de-identified data is generated.

5. The method according to claim 4, characterized in that, The step of determining the de-identification strategy for the entity based on the comprehensive confidence level corresponding to the entity, the preset parameters, the first threshold, and the second threshold includes: In response to the fact that the category corresponding to the entity belongs to a preset sensitive class, the product of the preset parameter and the second threshold is used as the first desensitization threshold; The noise threshold is determined based on the preset parameters and the first threshold. Based on the comprehensive confidence level of the entity, the first desensitization threshold, and the noise threshold, the desensitization strategy corresponding to the entity is determined.

6. The method according to claim 5, characterized in that, The step of determining the de-identification strategy corresponding to the entity based on the comprehensive confidence level, the first de-identification threshold, and the noise threshold includes: If the overall confidence level of the entity is greater than the first desensitization threshold, the preset first desensitization strategy is used as the desensitization strategy for the entity. In response to the overall confidence level corresponding to the entity being less than or equal to the first desensitization threshold, a desensitization strategy corresponding to the entity is determined based on the overall confidence level corresponding to the entity and the noise threshold.

7. The method according to claim 6, characterized in that, The step of determining the de-identification strategy for the entity based on the comprehensive confidence level and the noise threshold includes: If the overall confidence level of the entity is greater than the noise threshold, the preset second desensitization strategy will be used as the desensitization strategy for the entity.

8. The method according to claim 5, characterized in that, The step of determining the noise threshold based on the preset parameters and the first threshold includes: The fourth calculation result is obtained by subtracting the preset second value from the preset parameter. The fourth calculation result is multiplied by the first threshold to obtain the noise threshold.

9. A data processing apparatus, characterized in that, The device includes: An entity recognition module is used to acquire multiple image data and perform entity recognition based on the multiple image data to obtain multiple entities and the initial confidence level corresponding to each entity; An enhanced confidence determination module is used to, for each entity, in response to the initial confidence corresponding to the entity being less than a first threshold, perform enhanced processing on the image data corresponding to the entity in the multiple image data and then re-perform entity recognition to obtain the enhanced confidence corresponding to the entity; The comprehensive confidence level determination module is used to determine the comprehensive confidence level of the entity based on the initial confidence level and the enhanced confidence level corresponding to the entity. The de-identified data generation module is used to de-identify the entity based on the comprehensive confidence level corresponding to the entity, the first threshold, and preset parameters, and generate de-identified data.

10. A vehicle, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, when executing a program stored in memory, implements the method described in any one of claims 1-8.

Images